Recognition of the Martian minerals based on the deep belief networks method: Application in the CRISM images
-
摘要: 鉴于传统的光谱特征参数方法存在不能综合考虑光谱在整个波长范围内的谱形、对于单一吸收带相似的不同矿物难以区分等问题，研究采用深度置信网络方法对火星专用小型侦察影像频谱仪（CRISM）高光谱影像中的火星表面矿物进行自动识别，该算法具体包括：①预训练阶段。利用非监督算法逐层训练受限玻尔兹曼机，自动学习模型参数，提取光谱特征。②调优阶段。将自动学习的光谱特征输入分类器，采用反向传播算法对模型进行监督微调，识别矿物在CRISM影像中的分布。在算法的研究中，采用光谱比值方法降低火星表面灰尘等噪声对矿物光谱的影响，并探讨样本数量、隐含层节点数、网络深度等对算法识别精度的影响，试图构建适宜于CRISM影像火星表面矿物识别的深度置信网络模型。以火星表面镁铁蒙脱石和氯盐为例进行测试，实验结果表明：该方法能够对火星表面矿物进行自动识别，准确率达到85%以上，与光谱参数法的识别结果基本叠合，并能够探测光谱参数法未能识别的部分矿物分布。Abstract: In order to decrease recognition inaccuracies of the different minerals with similar single absorption peak by means of the spectral characteristic parameter methods which are difficult to estimate the spectrum of the whole wavelength range, this paper applied the deep belief networks (DBN) method to detect the Martian minerals from the hyperspectral images of the compact reconnaissance imaging spectrometer for Mars (CRISM). According to the method, firstly, the unsupervised layer-by-layer greedy algorithm is adopted to train each restricted Boltzmann machine (RBM) for the sake of learning parameters and extracting the spectral features of the minerals with a single bottom-up pass. Then, it takes advantage of the back propagation (BP) algorithm to tune the parameters learned in the train step and automatically identify the Martian minerals with coupling a suitable classifier. In this paper, the ratios of the minerals spectral and the dust spectral are utilized to identify the mineral samples for sake of decreasing the dust effect. Finally the influences of the sample size, the number of the hidden layer nodes, and the network depth are investigated to established the optimal deep belief networks for the recognition of the Martian minerals. As illustrated by the case of the Mg/Fe smectites and the chlorides from the CRISM images, the experimental results indicate that the recognition accuracy of the DBN method is more than 85%. In conclusion, the DBN method has a better performance in detecting some pixels of the minerals that the spectral parameter algorithm cannot detect in the CRISM images, and the deep learning method could be utilized in the recognition of the Martian minerals automatically.
-
Key words:
- recognition /
- deep belief networks /
- martian minerals /
- CRISM image
-
图 3 CRISM影像镁铁蒙脱石与氯盐光谱曲线示例
a.影像B001假彩色图(红色：2.38 μm; 绿色:1.80 μm; 蓝色:1.15 μm); b.影像AB81假彩色图(红色:D2300;绿色:ISLOPE1;蓝色:BD1900r2);光谱特征参数法提取镁铁蒙脱石与氯盐(B001); d.光谱特征参数法提取镁铁蒙脱石与氯盐(AB81); e.镁铁蒙脱石的光谱曲线; f.氯盐的光谱曲线; g.镁铁蒙脱石的比值光谱曲线; h.氯盐的比值光谱曲线; i.标准光谱库的氯盐和蒙脱石光谱曲线
Figure 3. Spectral curve samples of the Mg/Fe smectites and the chlorides from the CRSIM images
图 4 基于不同训练样本数量的分类结果
a.不同训练样本数量分类结果Kappa系数；b.不同训练样本数量分类结果整体精度、镁铁蒙脱石和氯盐的分类正确率，其中B001_OA、AB81_OA分别为影像B001、AB81中两类矿物总体分类精度，B001_SA、AB81_SA分别为影像B001、AB81中蒙脱石分类精度，B001_CA、AB81_CA分别为影像B001、AB81中氯盐分类精度
Figure 4. Maps showing the classification results based on the different sample sizes
图 5 基于不同隐含层节点数分类结果Kappa系数值
Figure 5. Map showing the Kappa coefficients based on the different node numbers of hidden layers
图 6 基于不同网络深度的分类Kappa系数
Figure 6. Kappa coefficients of the classification results based on the different network depth
图 8 基于深度置信网络模型影像AB81和B001中镁铁蒙脱石与氯盐的识别ROC曲线图
Figure 8. Maps showing the ROC curves of the classification results of the chlorides and the Mg/Fe smectites detected by the DBN method
图 9 基于自动编码机模型影像AB81和B001中镁铁蒙脱石与氯盐的识别ROC曲线图
Figure 9. Maps showing the ROC curves of the classification results of the chlorides and the Mg/Fe smectites detected by the autoencoder method
图 10 DBN方法与光谱特征参数法识别结果对比(影像AB81)
a.基于DBN方法的镁铁蒙脱石和氯盐分布图；b.基于光谱特征参数法提取的镁铁蒙脱石和氯盐分布图；c.基于DBN方法的镁铁蒙脱石分布图；d.基于光谱特征参数法的镁铁蒙脱石分布图；e.基于DBN方法的氯盐分布图；f.基于光谱特征参数法的氯盐分布图
Figure 10. Maps showing the comparison results of the DBN and the spectral characteristic parameter method (Image AB81)
图 11 DBN方法与光谱特征参数法识别结果对比(影像B001)
a.基于DBN方法的镁铁蒙脱石和氯盐分布图；b.基于光谱特征参数法提取的镁铁蒙脱石和氯盐分布图；c.基于DBN方法的镁铁蒙脱石分布图；d.基于光谱特征参数法的镁铁蒙脱石分布图；e.基于DBN方法的氯盐分布图；f.基于光谱特征参数法的氯盐分布图
Figure 11. Maps showing the comparison results of the DBN and the spectral characteristic parameter methods, respectively (Image B001)
图 12 由DBN方法识别的红圈和黄框标出的部分矿物光谱曲线
a.识别为镁铁蒙脱石的光谱曲线；b.识别为氯盐的光谱曲线
Figure 12. Maps showing a few spectral curves of the two minerals in the red circles and yellow rectangles detected by DBN method
表 1 评估镁铁蒙脱石和氯盐分布的光谱特征参数
Table 1. Spectral characteristic parameters for evaluating the distribution of Mg/Fe smectites and chlorides
名称 计算公式 矿物类型 D2300 $1 - \left( {\frac{{\frac{{R2290}}{{RC2290}} + \frac{{R2320}}{{RC2320}} + \frac{{R2330}}{{RC2330}}}}{{\frac{{R2120}}{{RC2120}} + \frac{{R2170}}{{RC2170}} + \frac{{R2210}}{{RC2210}}}}} \right)$
RC值为在1.8~2.53 μm之间反射率除以与斜率相应的值层状硅酸盐指示参数 ISLOPE1 $\frac{{R1815 - R2530}}{{W2530 - W1815}}$ 氯盐指示参数 BD1900R2 $1 - \frac{{\left( {\frac{{R1980}}{{RC1980}} + \frac{{R1914}}{{RC1914}} + \frac{{R1921}}{{RC1921}} + \frac{{R1928}}{{RC1928}} + \frac{{R1934}}{{RC1934}} + \frac{{R1941}}{{RC1941}}} \right)}}{{\left( {\frac{{R1862}}{{RC1862}} + \frac{{R1869}}{{RC1869}} + \frac{{R1875}}{{RC1875}} + \frac{{R2112}}{{RC2112}} + \frac{{R2120}}{{RC2120}} + \frac{{R2126}}{{RC2126}}} \right)}}$
RC值为在1.85~2.60 μm之间反射率除以与斜率相应的值H2O指示参数 注：R代表反射率，数字代表波长，如R1980表示1.98 μm处的反射率；W代表波段 
下载: 导出CSV
表 2 CRISM影像中镁铁蒙脱石和氯盐感兴趣区域
Table 2. ROI of the smectites and chlorides from CRISM images
名称 ID 分子 分母 中心像素位置 中心像素位置 名称 X Y ROI大小 名称 X Y ROI大小 B001_smec1 FRT0000B001 B001_semc1a 129 112 50 B001_smec1b 129 172 50 B001_smec2 FRT0000B001 B001_semc2a 61 374 48 B001_smec2b 61 117 45 B001_smec3 FRT0000B001 B001_semc3a 493 306 67 B001_smec3b 493 114 58 AB81_smec1 FRT0000AB81 AB81_semc1a 236 368 98 AB81_smec1b 236 71 105 AB81_smec2 FRT0000AB81 AB81_smec2a 340 369 42 AB81_smec2b 340 203 42 AB81_smec3 FRT0000AB81 AB81_smec3a 479 250 60 AB81_smec3b 479 49 65 B001_ch1 FRT0000B001 B001_ch1a 386 300 99 B001_ch1b 386 72 102 B001_ch2 FRT0000B001 B001_ch2a 254 281 63 B001_ch2b 254 58 60 B001_ch3 FRT0000B001 B001_ch3a 505 321 44 B001_ch3b 505 125 42 AB81_ch1 FRT0000AB81 AB81_ch1a 415 299 48 AB81_ch1b 415 32 50 AB81_ch2 FRT0000AB81 AB81_ch2a 309 263 49 AB81_ch2b 309 142 45
下载: 导出CSV
表 3 基于不同隐含层节点数分类结果
Table 3. Result details of classification based on the different node numbers of hidden layers
隐含层节点数 B001 AB81 Kappa系数 OA SA CA Kappa系数 OA SA CA 分类精度/% 分类精度/% 30 0.56 95.23 0.22 90.99 0.84 93.48 86.69 86.97 50 0.71 96.11 65.84 92.89 0.84 93.43 87.21 88.57 80 0.55 91.43 90.50 91.92 0.83 93.33 83.91 86.03 100 0.72 96.28 85.55 88.21 0.85 93.94 93.09 85.72 200 0.69 95.37 83.15 89.02 0.80 91.46 87.82 89.24 300 0.66 94.59 83.59 89.36 0.83 92.93 91.64 89.11
下载: 导出CSV
表 4 基于不同网络深度的分类结果
Table 4. Result details of classification based on the different network depth
网络深度 B001 AB81 Kappa系数 OA SA CA Kappa系数 OA SA CA 分类精度/% 分类精度/% 1 0.67 95.20 84.22 86.79 0.84 93.65 90.19 87.80 2 0.72 96.28 85.55 88.21 0.85 93.94 93.09 85.72 3 0.53 94.55 0.00 94.32 0.80 92.33 80.18 89.65 4 0.09 72.52 0.00 92.64
下载: 导出CSV
表 5 与自动编码机分类结果的对比
Table 5. Result details of classification compared with the autoencoder method
影像 算法 Kappa系数 OA SA CA S_RR C_RR S_AUC C_AUC 分类精度/% AB81 自动编码机 0.67 79.95 82.88 74.44 41.57 90.38 0.838 4 0.956 0 深度置信网络 0.85 93.94 93.09 85.72 90.10 91.50 0.960 8 0.965 0 B001 自动编码机 0.61 76.67 64.54 95.22 25.95 90.41 0.767 6 0.993 9 深度置信网络 0.72 96.28 85.55 88.21 85.55 90.20 0.877 2 0.902 5
下载: 导出CSV
-
[1] 赵斌魁, 孙平贺, 张绍和, 等."好奇"号火星探测器火星表面取样钻探近况[J].地质科技情报, 2018, 37(6):286-293. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=dzkjqb201806036 [2] Joshua A, Bandfiled L.Global mineral distributions on Mars[J].Journal of Geophysical Research, 2002, 107(6):1-20. http://core.ac.uk/display/10490750 [3] Christensen P R, Bandfield J L, Hamilton V E, et al.Mars global surveyor thermal emission spectrometer experiment:Investigation description and surface science results[J].Journal of Geophysical Research, 2001, 106(10):23823-23871. doi: 10.1029/2000JE001370/references [4] 祝民强, 周万蓬, 胡全一, 等.火星快车OMEGA高光谱探测矿物组成的新进展[J].地球科学进展, 2010, 25(7):691-697. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=dqkxjz201007003 [5] Poul F, Bilbring J P, Mustard J F, et al.Phyllosilicates on Mars and implications for early Martian climate[J].Nature, 2005, 438:623-627. doi: 10.1038/nature04274 [6] Ehlmann B L, Mustard J F, Swayze G A, et al.Identification of hydrated silicate minerals on Mars using MRO-CRISM:Geologic context near Nili Fossae and implications for aqueous alteration[J].Journal of Geophysical Research, 2009, 114(3):1-33. doi: 10.1029/2009JE003339/abstract [7] El-Maarry M R, Pommerol A, Thomas N.Analysis of polygonal cracking patterns in chloride-bearing terrains on Mars:Indicators of ancient playa settings[J].Journal of Geophysical Research, 2013, 118(11):2263-2278. doi: 10.1002/2013JE004463/citedby [8] Vivian-Beck C E, Seelos F P, Murchie S L, et al.Revised CRISM spectral parameters and summary products based on the currently detected mineral diversity on Mars[J].Journal of Geophysical Research Planets, 2014, 119(6):1403-1431. doi: 10.1002/2014JE004627 [9] Bornstein B, Castano R, Gilmore M S, et al.Creation and testing of an artificial neural network based carbonate detector for Mars recovers[C]//Cook K.2005 IEEE Aerospace Conference.MT, USA: IEEE Computer Society, 2005: 378-384. [10] Bornstein B, Castano R, Gilmore M S, et al.Generation and performance of automated jarosite mineral detectors for visible/near-infrared spectrometers at Mars[J].Icarus, 2008, 195(1):169-183. doi: 10.1016/j.icarus.2007.11.025 [11] Gilemore M S, Thompson D R, Anderson L J, et al.Superpixel segmentation for analysis of hyperspectral data sets, with application to compact reconnaissance imaging spectrometer for Mars data, Moon mineralogy mapper data[J].Journal of Geophysical Research Atmospheres, 2011, 116(7):4080-4093. https://www.researchgate.net/publication/251434186_Superpixel_segmentation_for_analysis_of_hyperspectral_data_sets_with_application_to_Compact_Reconnaissance_Imaging_Spectrometer_for_Mars_data_Moon_Mineralogy_Mapper_data_and_Ariadnes_Chaos_Mars [12] Carter J, Poulet F, Murchie S, et al.Automated processing of planetary hyperspectral datasets for the extraction of weak mineral signatures and applications to CRISM observations of hydrated silicates on Mars[J].Planetary and Space Science, 2013, 76(2):53-67. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=932774cb7ad4103a0de38b7aa793740d [13] 苏余斌, 詹云军, 黄解军, 等.面向高光谱矿物填图的多特征结合降维方法研究[J].地质科技情报, 2015, 34(5):206-211. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=dzkjqb201505032 [14] Melgani F, Bruzzone L.Classification of hyperspectral remote sensing images with support vector machines[J].IEEE Transactions on Geoscience and Remote Sensing, 2004, 42(8):1778-1790. doi: 10.1109/TGRS.2004.831865 [15] Fauvel M, Benediktsson J A, Chanussot J, et al.spectral and spatial classification of hyperspectral data using SVM sand morphological profiles[J].IEEE Transactions on Geoscience and Remote Sensing, 2008, 46(11):3804-3814. doi: 10.1109/TGRS.2008.922034 [16] Camps-Valls G, Bruzzone L.Kernel-based methods for hyperspectral image classification[J].IEEE Transactions on Geoscience and Remote Sensing, 2005, 43(6):1351-1362. doi: 10.1109/TGRS.2005.846154 [17] Samat A, Du P, Liu S, et al.Ensemble extreme learning machines for hyperspectral image classification[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2014, 7(4):1060-1069. doi: 10.1109/JSTARS.2014.2301775 [18] Hu Wei, Huang Yangyu, Wei Li, et al.Deep convolutional neural networks for hyperspectral image classification[J].Journal of Sensors, 2015, 2015(2):1-12. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=0525ad32bdb5ec66dbfadaf3cb6189d9 [19] Chen Yushi, Jiang Hanlu, Jia Xiuping, et al.Deep features extraction and classification of hyperspectral images based on convolutional neural networks[J].IEEE Transaction on Geosciences and Remote Sensing, 2016, 54(10):6232-6251. doi: 10.1109/TGRS.2016.2584107 [20] Li T, Zhang J P, Zhang Y, et al.Classification of hyperspectral image based on deep belief networks[C]//Caqnazzo M.2014 IEEE International Conference on Image Processing (ICIP).Paris: IEEE Computer Society, 2015: 5132-5136. [21] Zhao Wenzhi, Du Shihong.Spectral-spatial feature extraction for hyperspectral image classification:A dimension reduction and deep learning approach[J].IEEE Transaction on Geosciences and Remote Sensing, 2016, 54(8):4544-4554. doi: 10.1109/TGRS.2016.2543748 [22] Chen Yushi, Lin Zhouhan, Wang Gang, et al.Deep learning-based classification of hyperspectral data[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2014, 7(6):2094-2107. doi: 10.1109/JSTARS.2014.2329330 [23] 林杨挺.探索火星环境和生命[J].自然杂志, 2016, 38(1):1-7. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=zrzz201601002 [24] 徐丽坤, 刘晓东, 向小翠.基于深度信念网络的遥感影像识别与分类[J].地质科技情报, 2017, 36(4):244-249. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=dzkjqb201704032 [25] 李玮, 吴亮, 陈冠宇.基于遥感分类的深度信念网络模型研究[J].地质科技情报, 2018, 37(2):208-214. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=dzkjqb201802028 [26] Hinton G E, Osindero S.A fast learning algorithm for deep belief nets[J].Neural Computation, 2006, 18(7):1527-1554. doi: 10.1162/neco.2006.18.7.1527 [27] Bengio Y.Learning deep architectures for AI[J].Foundations and Trends in Machine Learning, 2009, 2(1):1-127. http://jamia.oxfordjournals.org/lookup/external-ref?access_num=10.1561/2200000006&link_type=DOI [28] Hinton G E.Training product of experts by minimizing contrastive divergence[J].Neural Computation, 2002, 14(8):1771-1800. doi: 10.1162/089976602760128018 [29] 刘超, 唐锡彬, 邓冬梅等.基于支持向量机回归的岩体变形模量预测[J].地质科技情报, 2018, 37(5):275-280. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=dzkjqb201805038 [30] Murchie S, Arvidson R, Bedini P, et al.Compact reconnaissance imaging spectrometer for Mar (CRISM) on Mars reconnaissance orbiter (MRO)[J].Journal of Geophysical Research, 2007, 112(5):431-433. doi: 10.1029/2006JE002682/abstract [31] Gurunadham R, Shashi Kumar.Extraction of aqueous minerals on Mars surface using CRISM based targeted reduced data records[M].Hyderabad:ISPRS, 2014. [32] 杨懿, 金双根, 薛岩松.利用CRISM数据探测火星表面含水矿物及其演化[J].深空探测学报, 2016, 3(2):187-194. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=sktcxb201602015 [33] McGuire P C, Bishop J L, Brown A J, et al.An improvement to the volcano-scan algorithm for atmospheric correction of CRISM and OMEGA spectral data[J].Planetary and Space Science, 2009, 57(7):809-815. doi: 10.1016/j.pss.2009.03.007 [34] Hinton G E.Reducing the dimensionality of data with neural networks[J].Science, 2006, 313:504-507. doi: 10.1126/science.1127647 -
下载:
点击查看大图
计量
- 文章访问数: 870
- PDF下载量: 1793
- 被引次数: 0
